|Publication number||US4216354 A|
|Application number||US 05/964,324|
|Publication date||Aug 5, 1980|
|Filing date||Nov 29, 1978|
|Priority date||Dec 23, 1977|
|Also published as||DE2860220D1, EP0002998A1, EP0002998B1|
|Publication number||05964324, 964324, US 4216354 A, US 4216354A, US-A-4216354, US4216354 A, US4216354A|
|Inventors||Daniel J. Esteban, Claude R. Galand, Daniel Mauduit, Jean Menez|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (2), Referenced by (102), Classifications (10)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This invention relates to voice coding at low bit rate or, more particularly, to digital transcoding to provide a compressed voice signal.
The various steps required for processing a time-varying signal which has to be recorded, transmitted or modified as required by the application, can be performed either in analog or digital mode. Analog processing usually involves, the use of complex and expensive means. Digital processing is generally preferred even though it involves a preliminary digital conversion of the signal to be processed. For this purpose, the signal is sampled, then quantized, and expressed in binary-coded words. Of course, it is obvious that the signal will have to be decoded at the end of the processing steps to put the information back into analog form useful to humans. The coding and decoding steps affect the information, usually by introducing noise, e.g., quantizing noise. However, by quantizing as accurately as possible, one can reduce quantizing noise. When this operation is carried out, all the samples, the amplitude of which lie between two consecutive digital values, are translated into the same number. This results in an inaccurate transcription which becomes more important as the number of the digital values which can be used for coding the signal is reduced. Therefore, it is necessary to increase this number to improve the coding quality. This, however, is undesirable under certain conditions since it may overload channels provided for transmitting the data, in the memories provided for storing the data and in the circuits provided for their processing, all of which affects the cost of the hardware used.
Many solutions have been proposed for improving the coding quality while reducing the quantizing bit number, however, these solutions are not complete and often appear to be expensive. It is therefore desirable to achieve a solution which tends to reduce the coding bit rate while maintaining the same quality or conversely to improve the coding quality at a given bit rate.
An object of this invention is to provide a process for improving the coding quality at a given total bit rate.
Another object of this invention is to provide a device for implementing said process.
It should be noted that the main portion of the signal processing operation required within the scope of this invention is to be performed in digital mode. This involves a preliminary signal sampling and quantization operation performed by using conventional methods using a bit rate high enough to reduce noise, and then recode the data at a lower bit rate. In other words, the information is compressed by performing a transcoding operation.
In addition, since the unique properties of the human voice will be used, it can be said that another object of this invention is to provide a voice compression process.
Another object of this invention is to provide a digital transcoder implementing said process.
The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of the preferred embodiments of the invention, as illustrated in the accompanying drawings.
Data compression is performed on a telephone bandwidth of a voice signal sampled and quantized with a relatively high number of bits, the frequency bandwidth of said voice signal is first split into high and low frequency bands. The high frequency band is processed to derive therefrom pre-emphasized parcor type of coefficients which are then requantized to generate COEF data. The low frequency band is split into p sub-bands and each sub-band is submitted to an adaptive sub-band block quantizing operation which generates SIGNAL data over the p sub-bands. High band energy information is also derived from said high frequency band, which energy information is quantized into ENERG data. The ENERG, SIGNAL and COEF data, together, provide data representing the compressed voice signal.
FIG. 1 is a block diagram of a coder constructed according to the invention.
FIGS. 2 to 6' illustrate various embodiments of the elements of the coder of FIG. 1.
FIG. 7 is a block diagram of a decoder constructed according to the invention.
FIG. 7' illustrates one of the elements of the decoder shown in FIG. 7.
FIGS. 8 and 9 illustrate various embodiments of some of the elements of the decoder shown in FIG. 7.
N-bit binary coding allows 2N different digital values to be defined. To perform this coding operation, the swing of the signal to be coded is split into 2N values. Consequently, the smaller the signal swing, the more accurate the coding.
Taking the characteristics of the voice signal into consideration, Chong Kwan Un and D. Thomas Magill have proposed in the IEEE Transactions on Communications, Vol. COM 23, No. 12, December 1975, to reduce the dynamic range of the voice signal before carrying out the coding step, by performing a deconvolution operation before coding the base band of the voice signal through a delta modulator. The deconvolution step involves the use of a self-adaptive digital filter. This filtering operation involves a large digital processing workload. In addition, with such an arrangement, it is more convenient to quantize differently the coefficients depending upon whether they are intended for the deconvolution operations or for being transmitted to a receiver device; which still further increases the computing power required to realize the coder-decoder assembly.
This invention provides a solution to the above mentioned problems. For this purpose, as shown in FIG. 1, the voice signal S(t) is filtered through FT in order to occupy only a typical telephone bandwidth (300-3400 Hz), then sampled at a first given rate 8 KHz and quantized in A/D at a relatively high number of bits (12 bits per sample). A/D provides a signal, the nth sample of which is designated Sn. (In the following, we can use the same reference for the signal itself and for its samples.) This signal is then transmitted to both a lowpass filter decimator (i.e., periodic resampler LPFD and to a high pass filter HPF. The combined bandwidth of these two filters can be made complementary to cover the above-indicated telephone bandwidth. The HPF output feeds a coefficient determining device CALC, the purpose of which will be explained later on. The data provided by the CALC is requantized in a quantizer QA2 to generate a signal designated COEF. The signal delivered by filter LPFD, the bandwidth of which extends up to 800 Hz only, is comprised of samples Xs provided at a frequency of 2 KHz. Samples Xs are requantized in quantizer SBC by blocks of K samples. This requantizing step involves the techniques described in United States patent application Ser. No. 895,204 filed Apr. 10, 1978, now U.S. Pat. No. 4,142,071, which claims priority of French patent application Ser. No. 77 13995 filed on Apr. 29, 1977 and provides digital data designated by SIGNAL. The output of filter decimator LPFD also provides information representing the energy of the high bandwidth of the input signal to be generated, which information is requantized in quantizer QA1, thus generating a signal designated by ENERG. When the transcoder of this invention is used in a digital transmission system, data ENERG, SIGNAL and COEF are multiplexed by a conventional multiplexor MPX, onto a single transmission line.
In a preferred embodiment of the invention, an analog filter FT having a bandwidth between 300 and 3400 Hz is used. The converter A/D by techniques well known in the art provides 12-bit linearly quantized samples Sn generated at a sampling frequency of 8 KHz.
FIG. 2 shows an embodiment of devices LPFD and H. B. Energy. Samples Sn are applied to two delay lines. DL1 which is twice as long as DL2 is part of a digital filter which also includes multipliers MULT and an adder Add1. These elements provide a lowpass transversal filter with a cut-off frequency of 800 Hz and of unity gain in the pass band. The frequency of samples Xs from LPFD is restored to 2 KHz by periodically resampling the signal at the output of the lowpass filter at a lower frequency rate, e.g. equal to 1/4 of the frequency of the samples fed into said low-pass filter one sample out of four at the output of the transversal filter. This so-called "decimation" operation is performed by using a switch SW1 periodically closed at a frequency of 2 KHz.
The information about the energy of the high frequency bandwidth of the voice signal ranging from 800 to 3400 Hz is obtained in a High-Band Energy measuring device designated by H. B. Energy by subtracting in SUBTR1, the samples from Add1 from those from DL2, then by squaring in SQ, the terms from SUBTR1 and thereafter by accumulating into ACC, the output of SQ for the duration of a sample block, i.e., a given period of time of 32 ms in this case. A signal RST is used to reset ACC to zero one or more times every 32 ms. The accumulator can also be replaced advantageously by a lowpass digital filter decimator, i.e., including a switch the switching frequency of which is adjusted in accordance with the frequency at which the corresponding energy information is to be used (1 to 4 times every 32 ms in this case). As required, the energy information can be requantized in circuit QA1 at a bit rate lower than the one of the terms from EN, in order to effect a reduction of the information bit rate.
Samples Xs from SW1 are requantized in an adaptive sub-band block requantizing means SBC. For this purpose, the bandwidth of samples Xs is split into sub-bands. Then each sub-band is subjected to a quantizing step which is dynamically adapted to the characteristics of the signal contained in said sub-band. The processes and devices used for this purpose are described in the above-identified U.S. patent application. In effect, it can be shown that bit rate ni to be assigned to the quantization of the ith sub-band to reduce the total quantizing noise on all the sub-bands is determined by the following equation: ##EQU1## where N represents the total number of bits for all the sub-bands,
Ei represents the sub-band energy contained in the ith sub-band when processing the corresponding block of samples,
p represents the number of sub-bands required for covering the whole spectrum of signal Xs.
Of course, since ni indicates a number of bits, only an approximate integer value of the result of operation (1) is used.
The quantizer adjustments are performed for the duration of a block, i.e., 32 ms in this case. (It is assumed that samples Sn have been taken from the coder input, in blocks of K=256 samples, at a frequency of 8 KHz). Therefore, during a time frame of 32 ms, it is possible to determine the amplitude limits within which the signal of each sub-band varies and since bit rate ni is known, to determine the value to which quantizing step Qi of the corresponding quantizer should be fixed. For instance, if it is assumed that the amplitude of the ith sub-band varies uniformly between two extreme values separated by ten volts and that the bit rate assigned to it is ni =1, a step Qi=5 volts should be selected. This means that all the samples exceeding 0 volt will be coded "1", the ones under 0 volt "0", or conversely.
When the sub-band splitting in the frequency domain is sufficiently sharp, the frequency spectrum in each channel is relatively decorrelated. Then, it is obvious that the sub-band energy is proportional to the square of the amplitude of the largest sample (M) (in absolute value) contained in it. Then, ##EQU2## λ being a proportionality coefficient.
By combining equations (1) and (2), one has: ##EQU3## From the above-given equations and by giving an approximate integer value of the computed results of ni, it is unlikely that one would obtain the following: ##EQU4## But when the various values of ni for all the sub-bands are obtained, they are readjusted to satisfy equation (4). This readjustment is carried out by a redistribution of bits, by adding or subtracting units to or from certain ni.
The preliminary analog/digital conversion in A/D has been carried out in a conventional way with a high bit rate N' to ensure a sufficient accuracy, then the device of this invention is used to requantize the signal samples to reduce, as required, the total bit rate used on all the sub-bands or alternatively to distribute the N available bits in a more rational way. Therefore, one defines: N<N'. (In certain cases, it would be possible to choose N=N', then there would not be any real compression, but a better bit allocation).
The device shown in FIG. 3 enables the above-described quantizing process to be carried out. In said device, samples Xs are transmitted to a bank of filters FB which splits the signal into p sub-bands containing samples Xi (with i=1, 2, . . . , p). Filter FB is constructed as shown in FIG. 4. In said filter, samples Xs are filtered through a system provided with a tree structure comprised of similar cells. Each cell includes two half-band quadrature filters (H1 and H2) and switches which select one out of two samples from each output of said filters. This is a decimation (periodic resampling) operation associated to the filtering operation performed through H1 or H2. For more details about this filtering operation, refer to IBM Technical Disclosure Bulletin, Vol. 19, No. 9, February 1977, pages 3438 and 3439. When p=4, the periodic resampling operation provides four sub-bands B1 through B4, and when the rate of Xs is fe =2 KHz, the samples are provided at a rate of 0.5 KHz over each of these sub-bands. In each sub-band, an energy measuring device Em sorts the samples it receives during a given period of time, to determine the largest one (M) in magnitude. In other words, (Em) determines values Mj for j=1, 2, . . . , p. Then these values are applied to a logarithm generator LOG which can be a table in which the coordinates of the points of function y=2x are recorded or to a microprocessor, the program of which performs the computation of the base 2 logarithm of Mj from a conventional limited development expression. Then values log Mj are applied to an adder SOM which adds said values log Mj, multiplied by 1/p, and subtracts the result of this operation from the predetermined value N/p. Therefore, the device SOM generates the term σ. At this point all of the information for quantizing the sub-bands are available. For a better understanding of the description, it is assumed here that each sub-band is quantized separately. Let us assume that DQi is the device quantizing the ith sub-band. As shown in FIG. 5, DQi includes: an adder ADDi, a quantizing step generator STi and a quantizer QTi. The addition of the values of log Mi and of σ in ADDi provides ni which is used in STi to generate the quantizing step Qi. ##EQU5## The information Qi being provided to the quantizer QTi, said quantizer QTi requantizes the samples Xi at bit Xi ≧0, QTi will provide Si =1 and when Xi <0, it will provide Si =0.
Thus, samples Xi initially coded in PCM mode at N' bits (12 bits for instance) are recoded at ni =1 bit.
By using the above described quantizer SBC, it is not only possible to save quantizing bits by selecting N<N' as desired, but also to ensure a more rational distribution of these quantizing bits.
As seen above, the operation of quantizer SBC requires the analysis of the sub-band contents over a predetermined period of time. Thus, it seems not desirable to change the bit rate ni and the quantizing step for each sample. On the other hand, this process is particularly adapted for the coding of samples by blocks as it is done in the Block Companded PCM type of coding process which can be briefly described as follows: for each block of samples, a scale factor C is chosen so that the largest sample of the block does not fall beyond the coding limits. Then the block samples are quantized. The assembly comprised of the scale factor (or block characteristic) and the quantized block of samples provides the digital information which fully defines the block. For more details about this type of coding one can refer to the article of A. Croisier concerning a lecture given at the Zurich International Seminar on Digital Communication of 1974. In this invention, ni will be thus defined for the duration of each block.
The value of Mi or a value close to it can be chosen as the scale factor Ci in which case equation (3) becomes: ##EQU6## Of course, when the signal is to be decoded so as to reconstitute the original information S(t), not only sequences Si of each channel will be needed but also scale factors Ci. If this type of coder-decoder is used in the communication domain, it is possible to reduce the chances for overloading the channels by not transmitting the bit rate information ni, since they can be recalculated at the receiver level from equation (6). Therefore, for the p sub-bands and for each time frame of 32 ms, p values of Qi, p values of Ci and the K values of the signal samples requantized by the devices of the p channels of the invention are transmitted.
In addition, while taking the characteristics of the signal to be quantized into account, it is possible to improve the coder performance by permanently forcing certain terms ni to zero, in particular the ones concerning the limiting sub-bands under 300 Hz and above 800 Hz (minimum and maximum values of i).
As shown in FIG. 1, samples Sn are also applied to a conventional high pass filter which eliminates the components of the signal under 800 Hz. Its output feeds device CALC which determines the coefficients characterizing the voice signal being processed. For a good understanding of the origin and function of these coefficients, one may refer to the article of L. S. Moye entitled, "Digital Transmission of Speech at Low Bit Rates" published in Vol. 47, No. 4 of the "Electric Communications Review" of 1972, page 217, and the following. The redundancy of the voice signal is shown in this article from which it appears that each sample of the voice signal can be estimated from a linear combination of previous samples by the following equation: ##EQU7## g being a predetermined integer. Various methods for determining the ai coefficients are known in the art, and in particular, the autocorrelation method of F. Itakura and S. Saito presented at a lecture given to the Japan Acoustical Association in 1969 and entitled, "Speech Analysis--Synthesis System Based on Partial Autocorrelation Coefficient. " This method tends to minimize the mean square error ER based on the following equation: ##EQU8## by performing a total estimation in which it is assumed that the signal is null outside the frame of the block to be processed. It is known in the art that ER is minimized when the following equation is satisfied: ##EQU9## By varying j and i, a series of linear equations is obtained in which the terms R represent the samples of the autocorrelation function of the voice signal. However, it is necessary to take into account that one has assumed for computing the correlation coefficients that the signal is null outside the analysis block. If this hypothesis is taken into account, one has: ##STR1##
This method is described in an article entitled, "Experimental Comparison Between Stationary and Non-Stationary Formulation of Linear Prediction Applied to Voiced Speech Analysis," published by Satish Chandra and Wen C. Lin in the IEEE Transactions on Acoustics, Speech and Signal Processing Review, Vol. ASSP22, No. 6 December 1974.
Starting from terms Rj, the parameters ai can be obtained in a recursive way by using an algorithm proposed by Levinson in the Journal of Mathematics and Physics, Vol. 25, No. 4, page 261, 1947. In fact, the algorithms of Levinson use so-called partial correlation coefficients k. Our device CALC computes these k coefficients. For this purpose, we use the methods and algorithms proposed by J. LeRoux and C. Gueguen in an article published in the IEEE Transactions on Acoustics, Speech and Signal Processing of June 1977.
In addition, it is possible to compensate for the combined effects of the glottal source and buccal cavity which induce a decrease of the voice signal spectrum at 6 db per octave, which is particularly perceptible in the case of voiced sounds by pre-emphasizing the signal before using it for calculating the autocorrelation coefficients. This pre-emphasis is accomplished by using a preemphasis filter having a z transfer function of the following type:
where α=R1/R0, R0 and R1 being the first two points of autocorrelation function of Sn. In this case, it is necessary to calculate the first terms of the autocorrelation function of the signal, to determine α=(R1/R0), then to filter the samples Sn by using a pre-emphasis filter H(z)=1-αz-1, and to recalculate the terms of the autocorrelation function. Filtering through H(z) will be avoided and thus, a portion of the total computing power required by the pre-emphasis and autocorrelation operations will be saved by using in CALC a digital filter of the type shown in FIG. 6. This filter is connected at the output of the device computing the terms R0, R1, and it derives the values of R'0 and R'1, . . . of the autocorrelation function, as if the signal applied to CALC was pre-emphasized. The remaining part of the LeRoux's method for determining coefficients ki remains unchanged. In other words, the samples Sn from filter HPF are applied to an autocorrelation function generator COMP (Ri) performing the operations of equation (8). The device COMP (Ri) includes a memory containing two consecutive blocks of data, and multiplying circuits, and computes the following: ##EQU10## By dividing R1 by R0, COMP(Ri) determines α. This coefficient is used in a circuit COMP (R'i) or preemphasis filter computing coefficients R'1 through the following equation:
R'i =Ri +α(-R.sub.|i-1| +αRi -R.sub.|i+1|).
Then, the terms R'i are applied to a partial autocorrelation circuit COMP (ki) generating the terms Ki defined by LeRoux and Guegen, which terms are defined here with respect to pre-emphasized autocorrelation coefficients R'1 and not with respect to terms R1. The device COMP (ki) can be realized by using a computer programmed for carrying out a recursive computation, the algorithm of which is represented on FIG. 6' (for g=3). The terms R' are loaded into adders (e0) forming a first stage. The value of ko is computed by using the contents of this first stage. The outputs of the adders of said first stage eventually multiplied by ko (as shown on the figure), feed a second stage of adders (e1) (see the figure) from the contents of which k1 is computed. The outputs of the stages of group (e1) eventually multiplied by k1, feed a third stage (e2), the contents of which allows k2 to be computed.
The coefficients ki will eventually be requantized in QA2 which can, as QA1, be of a conventional type but be provided with a quantizing step and characteristics which are redefined to be more closely adapted to the type of information to be processed, and to reduce the information rate. For instance, as far as QA2 is concerned, starting from statistical data, one has requantized terms ki with a number of bits fixed in accordance with the order i. The quantizer QA2 transmits a set of coefficients for each block of samples of the processed voice signal on channel COEF, for instance every 32 ms. Therefore, these coefficients are preemphasized parcor type coefficients.
The above indicated processing for determining ENERG, SIGNAL and COEF introduces delays. In effect, in the embodiment chosen as an example, terms Sn are taken by blocks of 256 samples. Therefore, for a sampling frequency of 8 KHz (Te=125 μs), the block has a duration of 32 ms. The delays due to each of filters HPF and LPFD are 28×125 μs. The one due to device CALC performing correlations is 32 ms. In order to have the terms COEF and SIGNAL in phase, it is therefore necessary that SBC introduce a delay of 32 ms. But the delay due to the SBC filter is 144×125 μs. Therefore, it is necessary to add a 112×125 μs delay line on the channel leading to the output SIGNAL. For the same reasons, a 256×125 μs delay line is needed in the path ENERG. (These delay lines are not shown on FIG. 1)
The above described schematics can be modified without departing from the scope of this invention. For instance, filter HPF may be eliminated in a simplified embodiment.
When the original voice signal, i.e., signal Sn, is to be reconstructed, it is necessary to recombine the ENERG, SIGNAL and COEF data. These operations are carried out in a receiver-decoder which is constructed as shown in FIG. 7. The data applied to the input line are, first of all, demultiplexed in DMPX which separates the data of channels ENERG, SIGNAL and COEF from each other.
The information SIGNAL is decoded in SBC which has to provide samples Xs. The device SBC shown in FIG. 8 includes four devices DQ1 to DQ4 (in this case p=4) feeding a set of filters carrying out the inverse of the operation performed by the set of filters used at the transmitter end and represented by FB in FIG. 3.
The device DQi processing the ith order sub-band receives requantized samples Si and information ni and Qi. It derives therefrom the information concerning bandwidth Bi as it appeared at the output of filter FB at the transmitter end (see FIG. 3). For this purpose, if it is assumed that ni =1, for instance, device DQi provides the digital value corresponding to Qi at its output when Si =1 and -Qi when Si =0. The outputs of circuits DQ1 to DQ4 are recombined into a filtering and interpolation assembly so as to reconstruct the original signal Xs, at 2 KHz. The filtering operation is performed by using elements H'1 and H'2 which consist of quadrature high-pass and low-pass half-band filters, respectively. The interpolation is ensured by interpolating a zero between two consecutive input samples, then by adding the outputs of each pair of filters taken so as to generate samples Xs at 2 KHz.
The Xs frequency is brought back to 16 KHz by an interpolator INT. This device computes the intermediate samples to be placed between two consecutive samples at 2 KHz, the interpolation can be achieved through a filter. The output path of signal Xs from the interpolator is split into a direct path and an indirect path. The indirect path includes a high band generator HB in which the low-band spectrum is spread by nonlinear distortion (in DIST), from there the high band is extracted by filtering through BPFD and said high band is modulated in EAD by the energy information provided by channel ENERG. In addition, since certain sounds such as the fricative sounds, for instance, provide a very low energy level in their low bandwidth, the energy of the signal provided by DIST can be insufficient to meet the requirements of the decoder of this invention. Therefore, we added energy, provided by a white noise generator WN, to the output of DIST. The level of this energy is to be adjusted so that it can ensure its functions for the sounds with low energy in the low frequency bandwidth without disturbing the system operation when processing voiced sounds. For this purpose, it is necessary that the white noise have a low amplitude with respect to high band energy of the voiced sounds. The WN adjustment can be either empirically or dynamically provided as will be seen later. The signal resulting from the addition of white noise to the information provided by DIST goes through bandpass filter decimator (BPFD) the low frequency cutoff of which is set to 800 Hz. The decimation operation performed through BPFD brings the sample frequency back to 8 KHz. These samples are fed into the EAD which provides information on the high frequency band of the voice signal received by normalizing the high band level. This normalization is ensured by measuring the energy E2 from bandpass filter decimator BPFD by comparing E2 with energy E1 from decoder DEC 1 placed on channel ENERG and finally modulating the signal from BPFD by the result of this comparison. DEC 1 performs the inverse of the operation performed in QA1, i.e., it provides the energy information expressed with the same units as the ones used in EN to obtain the actual high frequency band of the voice signal received by the decoder-receiver. The output of EAD is filtered through a transversal digital filter INVF, the coefficients of which are provided by channel COEF, which have been decoded in DEC 2. Device DEC 2 performs a dual operation: the first one is the inverse of the operation performed by QA2 to generate terms ki, the second one provides terms ai from terms ki. In practice, it is possible to avoid the conversion of terms ki into terms ai and to use directly terms ki as the coefficients of filter INVF provided that the latter is a lattice network designed in accordance with the teachings of J. D. Markel et al in their book entitled, "Linear Prediction of Speech," paragraph 5.4.
When terms ai are to be used, they may be recalculated by using the algorithm of FIG. 7' (the representation of which uses a symbolism similar to the one used in FIG. 6').
The reason for using a white noise source WN was mentioned above. It was also mentioned that it is possible to adjust empirically and in a fixed manner, the amplitude of the signals provided by said source. For this, one starts from a statistical and empirical evaluation of the energy level E2. But after having measured energy E2, it is also possible to use the result of this measurement to improve the adjustment of WN. For this purpose, we have in FIG. 9 another possible embodiment of HB, multiplier MU multiplying output WN by a coefficient (for instance, ranging from 1 to 10), provided by a table TB, the reading of which can depend either on E2, or on the relation between E1 and E2.
Finally, the original voice signal is reconstructed by adding the samples Xs from INT and brought back to 8 KHz in AD.
As indicated for the transmitter, the treatments applied to the various channels ENERG, SIGNAL and COEF introduce delays into the signals from INVF. Before recombining the information for regenerating the voice signal, it is necessary to compensate for these delays. This explains the presence of the 32 ms, 84×125 μs, 56×125 μs delay elements shown in FIG. 7. It is assumed for this that decoder SBC, interpolator INT and filter BPFD respectively, introduce delays of 144×125 μs, 28×125 μs and 28×125 μs.
In addition, it is possible to improve the performance of the receiver shown in FIG. 7 by modifying the ai or ki coefficients more than once per 32 ms. One should then determine, between two actually received sets of coefficients ai, intermediate coefficients calculated by linear interpolation while taking into account the sets of coefficients ai concerning the data blocks immediately preceding and following the block under process.
On the other hand, if the filter HPF is not used at the transmission end, it is necessary to place at the receiver end the adder AD at the input of filter INVF instead of putting it at the output of INV in which it receives its input from the two delay lines and provides its output to INVF and the output is taken directly from INVF.
While the invention has been particularly shown and described with reference to a preferred embodiment thereof, it will be understood by those skilled in the art that numerous changes in form and detail may be made therein without departing from the spirit and scope of the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US2949505 *||Aug 14, 1957||Aug 16, 1960||Bell Telephone Labor Inc||Reduced bandwith transmission system|
|US3952164 *||Jul 18, 1974||Apr 20, 1976||Telecommunications Radioelectriques Et Telephoniques T.R.T.||Vocoder system using delta modulation|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US4328395 *||Feb 4, 1980||May 4, 1982||Texas Instruments Incorporated||Speech synthesis system with variable interpolation capability|
|US4330689 *||Jan 28, 1980||May 18, 1982||The United States Of America As Represented By The Secretary Of The Navy||Multirate digital voice communication processor|
|US4359608 *||Aug 26, 1980||Nov 16, 1982||The United States Of America As Represented By The United States Department Of Energy||Adaptive sampler|
|US4360708 *||Feb 20, 1981||Nov 23, 1982||Nippon Electric Co., Ltd.||Speech processor having speech analyzer and synthesizer|
|US4386237 *||Dec 22, 1980||May 31, 1983||Intelsat||NIC Processor using variable precision block quantization|
|US4417102 *||Jun 4, 1981||Nov 22, 1983||Bell Telephone Laboratories, Incorporated||Noise and bit rate reduction arrangements|
|US4419540 *||Oct 5, 1981||Dec 6, 1983||Texas Instruments Incorporated||Speech synthesis system with variable interpolation capability|
|US4441200 *||Oct 8, 1981||Apr 3, 1984||Motorola Inc.||Digital voice processing system|
|US4441201 *||Jan 25, 1982||Apr 3, 1984||Texas Instruments Incorporated||Speech synthesis system utilizing variable frame rate|
|US4464782 *||Feb 8, 1982||Aug 7, 1984||International Business Machines Corporation||Transmission process and device for implementing the so-improved process|
|US4464783 *||Apr 20, 1982||Aug 7, 1984||International Business Machines Corporation||Speech coding method and device for implementing the improved method|
|US4516258 *||Jun 30, 1982||May 7, 1985||At&T Bell Laboratories||Bit allocation generator for adaptive transform coder|
|US4520499 *||Jun 25, 1982||May 28, 1985||Milton Bradley Company||Combination speech synthesis and recognition apparatus|
|US4535472 *||Nov 5, 1982||Aug 13, 1985||At&T Bell Laboratories||Adaptive bit allocator|
|US4538234 *||Sep 2, 1982||Aug 27, 1985||Nippon Telegraph & Telephone Public Corporation||Adaptive predictive processing system|
|US4541012 *||Jan 4, 1982||Sep 10, 1985||Compression Labs, Inc.||Video bandwidth reduction system employing interframe block differencing and transform domain coding|
|US4569075 *||Jul 19, 1982||Feb 4, 1986||International Business Machines Corporation||Method of coding voice signals and device using said method|
|US4589130 *||Jan 18, 1983||May 13, 1986||International Business Machines Corp.||Multirate digital transmission method and device for implementing said method|
|US4677671 *||Nov 18, 1983||Jun 30, 1987||International Business Machines Corp.||Method and device for coding a voice signal|
|US4692909 *||Jun 29, 1984||Sep 8, 1987||Amoco Corporation||Adaptive seismic signal processor|
|US4713776 *||May 16, 1984||Dec 15, 1987||Nec Corporation||System for simultaneously coding and decoding a plurality of signals|
|US4790015 *||Mar 25, 1987||Dec 6, 1988||International Business Machines Corporation||Multirate digital transmission method and device for implementing said method|
|US4802171 *||Jun 4, 1987||Jan 31, 1989||Motorola, Inc.||Method for error correction in digitally encoded speech|
|US4805193 *||Jun 4, 1987||Feb 14, 1989||Motorola, Inc.||Protection of energy information in sub-band coding|
|US4815134 *||Sep 8, 1987||Mar 21, 1989||Texas Instruments Incorporated||Very low rate speech encoder and decoder|
|US4831624 *||Apr 4, 1988||May 16, 1989||Motorola, Inc.||Error detection method for sub-band coding|
|US4845752 *||Oct 14, 1986||Jul 4, 1989||International Business Machines Corp.||Multi-signal processor synchronized system|
|US4890326 *||Mar 3, 1988||Dec 26, 1989||Rubiyat Software, Inc.||Method for compressing data|
|US4899384 *||Aug 25, 1986||Feb 6, 1990||Ibm Corporation||Table controlled dynamic bit allocation in a variable rate sub-band speech coder|
|US4907277 *||Jun 27, 1988||Mar 6, 1990||International Business Machines Corp.||Method of reconstructing lost data in a digital voice transmission system and transmission system using said method|
|US4920534 *||Feb 28, 1986||Apr 24, 1990||At&T Bell Laboratories||System for controllably eliminating bits from packet information field based on indicator in header and amount of data in packet buffer|
|US4941152 *||Aug 22, 1986||Jul 10, 1990||International Business Machines Corp.||Signal coding process and system for implementing said process|
|US4949383 *||Aug 21, 1988||Aug 14, 1990||Bristish Telecommunications Public Limited Company||Frequency domain speech coding|
|US4964166 *||May 26, 1988||Oct 16, 1990||Pacific Communication Science, Inc.||Adaptive transform coder having minimal bit allocation processing|
|US4979188 *||Apr 29, 1988||Dec 18, 1990||Motorola, Inc.||Spectrally efficient method for communicating an information signal|
|US5001758 *||Apr 8, 1987||Mar 19, 1991||International Business Machines Corporation||Voice coding process and device for implementing said process|
|US5067158 *||Jun 11, 1985||Nov 19, 1991||Texas Instruments Incorporated||Linear predictive residual representation via non-iterative spectral reconstruction|
|US5073938 *||Oct 17, 1989||Dec 17, 1991||International Business Machines Corporation||Process for varying speech speed and device for implementing said process|
|US5222189 *||Jan 29, 1990||Jun 22, 1993||Dolby Laboratories Licensing Corporation||Low time-delay transform coder, decoder, and encoder/decoder for high-quality audio|
|US5408580 *||Sep 21, 1992||Apr 18, 1995||Aware, Inc.||Audio compression system employing multi-rate signal analysis|
|US5517511 *||Nov 30, 1992||May 14, 1996||Digital Voice Systems, Inc.||Digital transmission of acoustic signals over a noisy communication channel|
|US5550949 *||Dec 23, 1993||Aug 27, 1996||Yozan Inc.||Method for compressing voice data by dividing extracted voice frequency domain parameters by weighting values|
|US5581651 *||Jul 5, 1994||Dec 3, 1996||Nec Corporation||Speech signal decoding apparatus and method therefor|
|US5606642 *||Sep 16, 1994||Feb 25, 1997||Aware, Inc.||Audio decompression system employing multi-rate signal analysis|
|US5627938 *||Sep 22, 1994||May 6, 1997||Lucent Technologies Inc.||Rate loop processor for perceptual encoder/decoder|
|US5657358 *||Apr 22, 1993||Aug 12, 1997||Interdigital Technology Corporation||Subscriber RF telephone system for providing multiple speech and/or data signals simultaneously over either a single or plurality of RF channels|
|US5687194 *||Apr 22, 1993||Nov 11, 1997||Interdigital Technology Corporation||Subscriber RF telephone system for providing multiple speech and/or data signals simultaneously over either a single or a plurality of RF channels|
|US5734678 *||Oct 2, 1996||Mar 31, 1998||Interdigital Technology Corporation||Subscriber RF telephone system for providing multiple speech and/or data signals simultaneously over either a single or a plurality of RF channels|
|US5737367 *||Oct 7, 1994||Apr 7, 1998||U.S. Philips Corporation||Transmission system with simplified source coding|
|US5752225 *||Jun 7, 1995||May 12, 1998||Dolby Laboratories Licensing Corporation||Method and apparatus for split-band encoding and split-band decoding of audio information using adaptive bit allocation to adjacent subbands|
|US5852604 *||May 20, 1996||Dec 22, 1998||Interdigital Technology Corporation||Modularly clustered radiotelephone system|
|US5870405 *||Mar 4, 1996||Feb 9, 1999||Digital Voice Systems, Inc.||Digital transmission of acoustic signals over a noisy communication channel|
|US6014374 *||Sep 9, 1997||Jan 11, 2000||Interdigital Technology Corporation||Subscriber RF telephone system for providing multiple speech and/or data signals simultaneously over either a single or a plurality of RF channels|
|US6208630||Dec 21, 1998||Mar 27, 2001||Interdigital Technology Corporation||Modulary clustered radiotelephone system|
|US6282180||Nov 4, 1999||Aug 28, 2001||Interdigital Technology Corporation|
|US6393002||Aug 6, 2001||May 21, 2002||Interdigital Technology Corporation|
|US6496488||Nov 2, 2000||Dec 17, 2002||Interdigital Technology Corporation||Modularly clustered radiotelephone system|
|US6757648||Jun 28, 2001||Jun 29, 2004||Microsoft Corporation||Techniques for quantization of spectral data in transcoding|
|US6771667||Feb 26, 2003||Aug 3, 2004||Interdigital Technology Corporation|
|US6842440||Apr 25, 2002||Jan 11, 2005||Interdigital Technology Corporation|
|US6954470||May 14, 2002||Oct 11, 2005||Interdigital Technology Corporation|
|US7069209||Jun 15, 2004||Jun 27, 2006||Microsoft Corporation||Techniques for quantization of spectral data in transcoding|
|US7082106||Jan 24, 2001||Jul 25, 2006||Multi-Tech Systems, Inc.||Computer-based multi-media communications system and method|
|US7082141||Jan 29, 2003||Jul 25, 2006||Multi-Tech Systems, Inc.||Computer implemented voice over data communication apparatus and method|
|US7092406||Jan 17, 2003||Aug 15, 2006||Multi-Tech Systems, Inc.||Computer implemented communication apparatus and method|
|US7092879||Jun 28, 2005||Aug 15, 2006||Microsoft Corporation||Techniques for quantization of spectral data in transcoding|
|US7245596||Jul 11, 2002||Jul 17, 2007||Interdigital Technology Corporation||Modularly clustered radiotelephone system|
|US7542555||Jan 25, 2006||Jun 2, 2009||Multi-Tech Systems, Inc.||Computer-based multifunctional personal communication system with caller ID|
|US7673321||Mar 2, 2010||Paul Yurt||Audio and video transmission and receiving system|
|US7730512||Apr 28, 2006||Jun 1, 2010||Acacia Media Technologies Corporation||Audio and video transmission and receiving system|
|US7818773||Nov 11, 2002||Oct 19, 2010||Acacia Media Technologies Corporation||Audio and video transmission and receiving system|
|US8032367 *||Jul 2, 2002||Oct 4, 2011||Nec Corporation||Bit-rate converting apparatus and method thereof|
|US8239190 *||Aug 7, 2012||Qualcomm Incorporated||Time-warping frames of wideband vocoder|
|US8270473||Jun 12, 2009||Sep 18, 2012||Microsoft Corporation||Motion based dynamic resolution multiple bit rate video encoding|
|US8311115||Jan 29, 2009||Nov 13, 2012||Microsoft Corporation||Video encoding using previously calculated motion information|
|US8396114||Mar 12, 2013||Microsoft Corporation||Multiple bit rate video encoding using variable bit rate and dynamic resolution for adaptive video streaming|
|US8457958||Jun 4, 2013||Microsoft Corporation||Audio transcoder using encoder-generated side information to transcode to target bit-rate|
|US8705616||Jun 11, 2010||Apr 22, 2014||Microsoft Corporation||Parallel multiple bitrate video encoding to reduce latency and dependences between groups of pictures|
|US8935156||Apr 15, 2014||Jan 13, 2015||Dolby International Ab||Enhancing performance of spectral band replication and related high frequency reconstruction coding|
|US9218818||Apr 27, 2012||Dec 22, 2015||Dolby International Ab||Efficient and scalable parametric stereo coding for low bitrate audio coding applications|
|US9245533||Dec 9, 2014||Jan 26, 2016||Dolby International Ab||Enhancing performance of spectral band replication and related high frequency reconstruction coding|
|US9245534||Aug 19, 2013||Jan 26, 2016||Dolby International Ab||Spectral translation/folding in the subband domain|
|US20030006916 *||Jul 2, 2002||Jan 9, 2003||Nec Corporation||Bit-rate converting apparatus and method thereof|
|US20040225506 *||Jun 15, 2004||Nov 11, 2004||Microsoft Corporation||Techniques for quantization of spectral data in transcoding|
|US20050232497 *||Apr 15, 2004||Oct 20, 2005||Microsoft Corporation||High-fidelity transcoding|
|US20050240398 *||Jun 28, 2005||Oct 27, 2005||Microsoft Corporation||Techniques for quantization of spectral data in transcoding|
|US20080052065 *||Aug 22, 2006||Feb 28, 2008||Rohit Kapoor||Time-warping frames of wideband vocoder|
|US20090251829 *||Apr 2, 2008||Oct 8, 2009||Headway Technologies, Inc.||Seed layer for TMR or CPP-GMR sensor|
|US20090326935 *||Jul 7, 2007||Dec 31, 2009||Head Inhimillinen Tekiji Oy||Method of treating voice information|
|US20100189179 *||Jan 29, 2009||Jul 29, 2010||Microsoft Corporation||Video encoding using previously calculated motion information|
|US20100189183 *||Jan 29, 2009||Jul 29, 2010||Microsoft Corporation||Multiple bit rate video encoding using variable bit rate and dynamic resolution for adaptive video streaming|
|US20100316126 *||Jun 12, 2009||Dec 16, 2010||Microsoft Corporation||Motion based dynamic resolution multiple bit rate video encoding|
|US20140012571 *||Aug 1, 2013||Jan 9, 2014||Huawei Technologies Co., Ltd.||Method and apparatus for providing signal processing coefficients|
|USRE39080||Aug 13, 2002||Apr 25, 2006||Lucent Technologies Inc.||Rate loop processor for perceptual encoder/decoder|
|USRE40280||Oct 12, 2005||Apr 29, 2008||Lucent Technologies Inc.||Rate loop processor for perceptual encoder/decoder|
|USRE40281 *||Nov 23, 2004||Apr 29, 2008||Aware, Inc.||Signal processing utilizing a tree-structured array|
|WO1984004194A1 *||Mar 12, 1984||Oct 25, 1984||American Telephone & Telegraph||Speech pattern processing utilizing speech pattern compression|
|WO1989009965A1 *||Mar 16, 1989||Oct 19, 1989||Motorola, Inc.||Error detection method for sub-band coding|
|WO1989010661A1 *||Apr 3, 1989||Nov 2, 1989||Motorola, Inc.||Spectrally efficient method for communicating an information signal|
|WO1989011718A1 *||May 25, 1989||Nov 30, 1989||Pacific Communication Sciences, Inc.||Improved adaptive transform coding|
|WO1994007237A1 *||Sep 10, 1993||Mar 31, 1994||Aware, Inc.||Audio compression system employing multi-rate signal analysis|
|WO1994012932A1 *||Nov 29, 1993||Jun 9, 1994||Digital Voice Systems, Inc.||Coding with modulation, error control, weighting, and bit allocation|
|U.S. Classification||704/229, 704/217, 704/212, 704/230, 704/205|
|International Classification||H04B1/66, H03M7/30, H04B14/04|